SuperSonic Imagine Ultrafast Imaging & ultrafast compound doppler imaging providing full blood flow characterization

1. 134 IEEE TRANSACTIONS ON ULTRASONICS, FERROELECTRICS, AND FREQUENCY CONTROL , . 58, . 1, JANUARY 2011
Ultrafast Compound Doppler Imaging:
Providing Full Blood Flow Characterization
Jeremy Bercoff, Gabriel Montaldo, Thanasis Loupas, David Savery, Fabien Mézière, Mathias Fink,
and Mickael Tanter
Abstract—Doppler-based flow analysis methods require ac- velocity as a function of time, the mean flow velocity as
quisition of ultrasound data at high spatio-temporal sampling a function of time, the resistance and pulsatility indices
rates. These rates represent a major technical challenge for
ultrasound systems because a compromise between spatial and
within a cardiac cycle, the spectral broadening index, etc.
temporal resolution must be made in conventional approach- [1]. Spectral Doppler analysis requires continuous acquisi-
es. Consequently, ultrasound scanners can either provide full tions or very high sampling rates (several thousand hertz).
quantitative Doppler information on a limited sample volume Flow quantification is then typically available only at a
(spectral Doppler), or averaged Doppler velocity and/or power single location (sample volume) or multiple locations along
estimation on a large region of interest (Doppler flow imaging).
In this work, we investigate a different strategy for acquiring
the same line (multigating). Color flow imaging overcomes
Doppler information that can overcome the limitations of the the limited spatial sampling of the spectral analysis by
existing Doppler modes by significantly reducing the required partly sacrificing the quantitative analysis, by reducing
acquisition time. This technique is called ultrafast compound the observation time at any given location and spreading
Doppler imaging and is based on the following concept: instead the ultrasound firings over a 2-D region of interest. The
of successively insonifying the medium with focused beams,
several tilted plane waves are sent into the medium and the
information displayed is the mean flow velocity and/or
backscattered signals are coherently summed to produce high- Doppler power estimated over an extended area. Those
resolution ultrasound images. We demonstrate that this strat- modes are displayed in real time at frame rates that are
egy allows reduction of the acquisition time by a factor of up usually around a few hertz.
to of 16 while keeping the same Doppler performance. Depend- The big challenge of Doppler modes arises from the fact
ing on the application, different directions to increase perfor-
mance of Doppler analysis are proposed and the improvement
that physicians ideally require simultaneous real time dis-
is quantified: the ultrafast compound Doppler method allows play of B-mode (gray scale) and PW-mode (duplex mode),
faster acquisition frame rates for high-velocity flow imaging, or or even B-, color- and PW-modes (triplex mode). Duplex
very high sensitivity for low-flow applications. Full quantitative and triplex simultaneous modes have become standard on
Doppler flow analysis can be performed on a large region of ultrasound systems, but suffer from frame rate limitations
interest, leading to much more information and improved func-
tionality for the physician. By leveraging the recent emergence
in deep organs such as the liver or heart. Duplex and tri-
of ultrafast parallel beamforming systems, this paper demon- plex modes represent major technical challenges because
strates that breakthrough performances in flow analysis can be they require complex sequencing, high-energy ultrasound
reached using this concept of ultrafast compound Doppler. transmission, and high processing power. Severe tradeoffs
on imaging mode quality and/or frame rate are necessary.
Consequently, there is a crucial need to significantly re-
I. I duce the number of ultrasound firings required to perform
Doppler analysis (i.e., reduce the acquisition time) while
D - imaging methods are well-established
tools on ultrasound systems for flow analysis and
quantification, and have become mandatory in the con-
keeping constant or increasing performance.
Academic research into overcoming this issue has been,
and continues to be, extensive. Many directions have been
text of cardiovascular disease assessment as well as cancer considered. The simplest solution consists of reducing the
diagnosis. There are two different kinds of Doppler modes number of transmit beams per color flow image by widen-
available: spectral analysis (continuous or pulsed) and col- ing them [3]. Such an approach is currently implemented
or-coded flow velocity and/or power imaging [1], [2]. on ultrasound systems but requires tradeoffs between sen-
Spectral analysis Doppler offers excellent temporal res- sitivity and resolution to obtain a significant reduction
olution and provides in-depth quantification of flow char- of the acquisition time. Other approaches have been pro-
acteristics by means of quantities such as the peak flow posed, such as performing simultaneous transmissions [4]
and parallel beamforming, using synthetic aperture imag-
Manuscript received June 19, 2010; accepted October 8, 2010. ing [5], [6], or reducing the number of samples required to
J. Bercoff, T. Loupas, D. Savery, and F. Mézière are with Super- perform the Doppler estimation (ensemble length) while
Sonic Imagine, R&D, Aix en Provence, France (e-mail: jeremy.bercoff@ introducing higher-performance processing methods [7],
supersonicimagine.fr).
G. Montaldo, M. Fink, and M. Tanter are with Institut Langevin, designing pulses able to perform B- and Doppler-mode
École Supérieure de Physique et de Chimie Industrielles de la Ville de imaging simultaneously [8]. Such approaches often require
Paris (ESPCI) ParisTech, Centre National de la Recherche Scientifique the use of open and fully programmable electronic plat-
(CNRS), Institut National de la Santé et de la Recherche Médicale (IN-
SERM), Paris, France. forms [9], [10]. Although proposed solutions show prom-
Digital Object Identifier 10.1109/TUFFC.2011.1780 ising results, they add complexity to the Doppler-mode
0885–3010/$25.00 © 2011 IEEE

2.   .:   D  135
sequence and processing paths. As a consequence, most directions (or lines). Each image line is then computed
of them have not yet become standards in current ultra- by processing the backscattered echoes coming from the
sound systems. insonified direction. The maximal frame rate to produce a
In previous work, we proposed the use of plane-wave focused image is set by the following equation:
insonifications to perform Doppler-based tissue motion
analysis [11]. Plane-wave transmission represents the most c 1
Ffoc , (1)
efficient solution in terms of number of firings because the 2Z n Lines
whole medium is insonified in one shot. Ultrafast frame
rates (several thousands of hertz) can therefore be achieved, where Z is the maximal depth of the image, c is the speed
and this has led to the introduction of a new quantitative of sound, and nLines is the number of insonified lines. De-
elasticity imaging mode [12]. The plane wave technique pending on the application and the depth of interrogation,
implies compromises among resolution, contrast, and sen- frame rates varying from a few tens of hertz down to a few
sitivity that are not significant for tissue motion analysis hertz are typically achieved.
but may become important when dealing with weak blood Ultrafast imaging can be performed by insonifying the
flow scatterers. Udesen et al. [13] recently tested the plane medium with a single plane wave transmit. Backscattered
wave technique for color flow imaging. Coded excitations echoes are simultaneously recorded from the entire scan
were used to improve the signal-to-noise ratio, but, as plane, and all imaging lines are simultaneously computed
stated by the authors, the poor contrast of the technique using parallel beamforming processes. In this case, the
limits its application to flow analysis in large arteries. maximal frame rate is [12], [16]
A way to improve the performance of ultrafast plane
wave imaging is to use several tilted plane waves [14].
c
These waves are sent into the medium and the backscat- Fflat . (2)
tered signals are coherently summed to produce a fully 2Z
dynamically focused image (in transmit and receive). Re-
cently, we introduced a new imaging method based on this The plane wave imaging method is the most efficient way
approach called the ultrafast plane wave compound tech- to increase frame rate, at the expense of image contrast
nique [15]. We demonstrated that this technique allows and spatial resolution [16], [17].
the realization of a B-mode of equivalent quality to the In the ultrafast compound imaging method, a set of
standard focused approach with one-third the number of plane waves (NAngles) are sent into the medium at dif-
insonifications. We also successfully applied this concept ferent angles at an ultrafast frame rate. The backscat-
to shear-wave-based elastography, allowing improvement tered echoes are, in a first processing step, beamformed
of this mode in terms of resolution, contrast, and sensitiv- to produce NAngles ultrasound images. Each ultrasonic
ity. image is produced by applying a conventional dynamic
This paper investigates the ultrafast plane wave com- receive focusing along each line of the image (conventional
pound technique in the framework of Doppler-based flow delay-and-sum technique, fixed aperture ratio F/D ~ 1).
analysis methods. The new technique is called ultrafast In a second step, these beamformed images are coherently
compound Doppler imaging and is described in Section II. summed to obtain a compounded image which is dynami-
Section III evaluates and quantifies the performance of the cally focused in transmit and receive. It is important to
new color flow imaging mode in phantoms and compares note that the summation is done coherently before any
it to conventional focused color flow imaging. It is shown nonlinear process (envelope detection, etc.). The frame
that, for a given mode performance, the acquisition time rate is, in this case:
can be reduced by a factor of up to 16. Based on these re-
sults, Section IV demonstrates how color flow imaging can c 1
Fcomp . (3)
be enhanced using this insonification strategy through im- 2Z N Angles
proved sequencing and processing schemes. In vivo results
are presented. Section V proposes new tools for displaying Compared with a single flat insonification, the ultra-
and analyzing the flow data provided by ultrafast imaging fast compound imaging frame rates are reduced by NAngles
(>500 Hz). Finally, Section VI discusses real-time imple- (number of plane wave angles) to improve image quality
mentation of the new mode on an ultrasound system, and (contrast, resolution). In a previous article [15], it was
Section VII summarizes the conclusions of this study. demonstrated that when using NAngles ≈ 40, the ultrafast
compound imaging method has resolution, contrast, and
signal-to-noise ratio equivalent to the conventional focused
II. B method. The acquisition time is then reduced by a typical
factor of 3 to 6, depending on the number of focused lines
A. Ultrafast Compound Imaging (128 to 256) used in the conventional method. The con-
cept of plane wave compounding for increased image qual-
In conventional ultrasound imaging, the medium is se- ity has been successfully applied to the field of transient
quentially insonified using focused beams along different elastography [15]. This work investigates the performance

3. 136 IEEE TRANSACTIONS ON ULTRASONICS, FERROELECTRICS, AND FREQUENCY CONTROL , . 58, . 1, JANUARY 2011
of the ultrafast compound imaging method for Doppler-
based flow analysis.
B. New Technological Implementation for Fully
Parallel Beamforming
Ultrafast compound imaging requires the ability to ac-
quire and process ultrasound images at very high frame
rates, typically thousands of hertz. Conventional ultra-
sound systems usually reach frame rates of a few tens
of hertz as medium insonification and signal processing
are serialized by image line. Implementation of ultrafast
compound approaches is therefore not possible on such
systems because they require full parallelization of the im-
age formation process.
New platform architectures are needed, based on the com-
bination of ultrafast raw RF data acquisition capabilities and
full software-based and parallelized beamforming schemes.
The platform used in this work (Aixplorer, SuperSonic Imag-
ine, Aix en Provence, France) meets those requirements and
enables implementation of such new schemes.
C. Conventional Color Flow Imaging
In color flow imaging, flow velocity estimation relies on
the use of N narrowband (a few cycles) transmit pulses
sent at a fixed pulse repetition frequency (PRF) to esti-
mate the Doppler frequency (Fs). N is commonly referred
as the ensemble length. Based on the Nyquist theorem,
Fig. 1. (a) Conventional color mode and (b) ultrafast compound Doppler
to avoid aliasing, the PRF must be at least two times the sequences.
highest flow-related Doppler frequency of interest:
PRFflow 2 Fs . (4)
D. Ultrafast Compound Doppler Imaging
The main steps of the processing are wall filtering to dis-
In the ultrafast compound approach, a set of NAngles
criminate tissue echoes from the flow signal and veloc-
tilted plane waves are transmitted at the ultrafast frame
ity estimation, most commonly based on autocorrelation
rate (PRFmax) to generate a compounded image, as il-
methods [18]–[20].
lustrated in Fig. 1. The sequence is repeated N times at
In this conventional approach, the sequencing strat-
PRFflow to be able to perform wall filtering and flow veloc-
egy is determined by the ratio between the maximum
ity estimation.
PRF achievable by the system at the considered depth
The maximum number of angles is determined by the
(PRFmax) and the necessary PRF to detect the desired
same parameter that controls the number of segments in
maximum flow velocity (PRFflow). This ratio gives the
the conventional method: the ratio between PRFmax and
number of lines that can be sequentially insonified before
PRFflow:
going back to the first insonified line:
NAnglesMax = PRFmax/PRFflow. (7)
Nlines = PRFmax/PRFflow. (5)
One obvious advantage of the ultrafast compound approach
To generate a color flow image that contains more lines
is that the concept of scanning the color image on a segment-
than Nlines, the image is subdivided into several segments
by-segment basis disappears. Because the whole medium is
of Nlines and the color sequence and processing are done
insonified for each transmision, there are no more tradeoffs
sequentially for all segments as illustrated in Fig. 1(a).
between frame rate and size of the color box caused by se-
The number of firings necessary to compute a full color
quence timing issues. Moreover, flow velocity estimation is
flow image is given by the following formula
performed simultaneously for all pixels and not at different
NFiringsC = Nlines · NSegments · N, (6) time instances, as in the conventional approach, leading to
true 2-D real-time Doppler flow imaging.
where NSegments is the number of segments needed to com- The number of firings necessary for a full color flow im-
plete the full color image. age is no longer linked to the number of lines within the

4.   .:   D  137
Fig. 2. Experimental 2-D PSFs of (a) focused, (b) flat, and (c) ultrafast compound (with 9 angles) methods. (d) A transverse cut of the PSFs at
the scatterer depth.
image, but to the number of tilted plane waves transmit- section to provide insights for the next color flow section.
ted, according to All experiments were conducted on ultrasound phantoms
using the Aixplorer ultrasound system (SuperSonic Imag-
NFiringsUltrafast = NAngles · N. (8)
ine). The probe used is a standard linear probe (128 ele-
The gain GAT in acquisition time between conventional ments, 0.3 mm pitch, 5 to 12 MHz) dedicated to small
and ultrafast acquisitions is then set by the following for- parts and vascular applications. The probe was driven
mula: by the system at ±50 V. To perform image comparisons,
G AT N FiringsC/N FiringsUltrafast the following parameters were chosen: Transmit pulse for
(9) both methods: 3 cycles at 5 MHz, f-number for the focused
(N lines N N Segments)/(N Angles N).
method = 3, maximal angle values for the compounded
approach: ±9°. The maximum angle affects the resolu-
According to (5) and (7), and considering NAngles = tion of the compounded image, and has been chosen to
NAnglesMax = Nlines, this leads to the gain match the resolution of the focused method (f-number =
3). The number of angles used in the compound plane
GAT = NSegments. (10)
wave method is a varying parameter of our experiments.
For high-speed flows, the typical number of segments in Signals received by the system were sampled at 20 MHz.
a conventional color image is quite large, reaching up to Ultrasound images were computed with a wavelength res-
NSegments = 64 or even more (very high PRFFlow and large olution of 0.3 mm.
color box). Therefore, the acquisition time gain of the Spatial resolution and contrast were calculated from
ultrafast compound method is potentially huge for such the experimental point spread function (PSF) of each im-
large color boxes and high PRFs. aging sequence on a single strong scatterer (50-µm wire
For low-speed flows, the gain in acquisition time is not immersed in water). The 2-D PSFs for both modes were
important because NSegment is small (typically NSegment = calculated and compared with the so-called Flat mode,
1 to NSegment = 3). However, the gain in terms of sensitiv- which corresponds to the transmission of a single, un-
ity is considerable, because each pixel is insonified NAngles steered, plane wave. An example of 2-D PSFs for both
times more when using ultrafast compound Doppler. methods is shown in Fig. 2. The lateral resolution is as-
To evaluate the relevance of the ultrafast compound sessed by measuring the width of the PSF at the −6-dB
method from a practical point of view, its performance level. The axial resolution corresponds to the dimension
shall be quantified and compared with the conventional of the PSF at the −6-dB level in the axial (depth) direc-
focused Doppler technique. The next section is dedicated tion. Finally, the anechoic contrast is calculated as the
to the in vitro assessment of the ultrafast compound Dop- ratio between energy outside a circle of 5λ centered in the
pler mode performance using the conventional color flow PSF and the energy of the complete PSF (λ being the
mode as a reference. wavelength corresponding to the central frequency of the
pulse).
III. U C D I:
P A A. Resolution
The performances obtained in terms of resolution and The lateral resolution obtained for the three acquisi-
contrast using plane wave compounding are studied in this tion methods is shown in Table I. Although the single

5. 138 IEEE TRANSACTIONS ON ULTRASONICS, FERROELECTRICS, AND FREQUENCY CONTROL , . 58, . 1, JANUARY 2011
where NAngles is the number of angles, g is the antenna
gain, ZF is the focal depth, λ is the wavelength, and D is
the array aperture. Considering a typical case in which λ
= 0.3 mm, ZF = 30 mm, and D = 9 mm, the SNR should
be the same for both methods when the number of angles
is equal to 9. If using more angles, the plane wave com-
pounding method should provide better SNR than the
conventional approach.
Experiments were conducted on a phantom to confirm
these findings. The SNR is measured using a phantom
containing a homogeneous distribution of scatterers. This
phantom is placed on a vibration-free table and a large
set of images is acquired. For each pixel of the image, the
Fig. 3. Anechoic contrast for different methods. Using 9 angles, the con- mean signal s(x, z) and its standard deviation is calculated
trast is only 5 dB higher than in the focused mode. using the complete set of acquisitions.
These values depend mainly on depth. To obtain a pre-
cise profile of the SNR as a function of depth, the SNR
flat mode presents a lower lateral resolution, the ultra- values are averaged in the lateral direction and the final
fast compound method has equivalent resolution to the measured SNR is
focused mode. This was expected, because the lateral
resolution depends on the value of the maximum angles
s(x, z )
chosen (not on the number of angles used) [15]. These SNR(z ) x
. (12)
angles (±9°) have been effectively chosen to match the x
(x, z )
focused mode apertures and resolution. The axial resolu-
tion is also shown in Table I. Because it depends only on Fig. 4(a) shows the depth dependence of the SNR for the
the bandwidth of the ultrasonic pulse, it is almost identi- conventional focused method and an ultrafast compound
cal for all methods. sequence relying on 16 angles. The focused method has a
lower SNR than the compound except at focal depth.
B. Anechoic Contrast To compare both methods, we can define the SNR gain
as SNRcomp/SNRfoc. In Fig. 4(b), one can see that this
The anechoic contrast versus the number of angles is gain varies from 10 to 0 dB with a mean 5 dB gain over
shown in Fig. 3. The contrast decreases rapidly with the all depths. Using an ultrafast compound sequence of 9
number of angles: using NAngles = 9, the contrast level is angles, the mean gain across the whole image is reduced
at −37 dB, only 5 dB higher than that for the focused to approximately 2.5 dB.
method. For NAngles = 16, the contrast difference is only
2 dB. As we will see in the next section, those numbers of
angles are an excellent choice for low-flow velocity imaging D. Flow Analysis
that requires high sensitivity and resolution.
Experiments were conducted on a calibrated Doppler
C. Signal-to-Noise Ratio phantom (ATS 523A, ATS Laboratories Inc., Bridgeport,
CT) with blood mimicking fluid (Shelley Medical Imag-
Montaldo et al. studied the signal-to-noise ratio of the ing Technologies, London, Ontario, CA) circulating with
synthetic image obtained using compounded plane wave a mean flow velocity of 4 cm/s in a 4-mm-diameter ves-
insonifications compared with conventional B-mode im- sel. The linear probe and acquisition parameters were the
ages [15]. Assuming independent noise between insonifica- same as those used previously. The acquisition parameters
tions, it was shown that the SNRs of both imaging meth- are summarized in Table II. For the focus method, the
ods (at the focal depth of the conventional focused image) beam focus has been adjusted in the middle of the vessel
are linked by the following relation for the reference central line of the box (40 mm absolute
value)
SNR Comp N Angles N AnglesZ F Figs. 5(a) and 5(b) compare the power Doppler images
, (11)
SNR Foc g D for both methods. Figs. 5(c) and 5(d) compare the color
TABLE I. S R   D M.
Compound Compound
Resolution Focused Flat 9 angles 16 angles
Axial (mm) 1.07 1.10 1.01 1.02
Lateral (mm) 0.54 0.86 0.53 0.53

6.   .:   D  139
Fig. 5. Comparison between focused and ultrafast compound Doppler
images in a vessel phantom. (a) and (b) Power Doppler for focused
and compound modes, respectively (scale in decibels). (c) and (d) Color
Doppler for focused and compound method, respectively (scale in centi-
meters/second).
Fig. 4. (a) SNR versus depth: The ultrafast compound method with 16
angles has a higher SNR than the focused except at focal depth, where
they are practically identical. (b) SNR gain versus depth. The mean gain flow imaging and ultrafast plane wave compounding, re-
is approximately 5 dB. spectively. Hence, it is clearly demonstrated that the tis-
sue clutter level is similar for both imaging methods.
For the calculation of the MSE, the theoretical Poi-
Doppler images in the same configuration. Qualitatively, seuille profile v = vmax[1 − (r/R)2] is defined, where R is
the methods are very similar. the tube radius, r is the radial position within the tube,
For a quantitative comparison of both Doppler ap- and vmax is the maximal velocity in the center of the tube.
proaches, two quantities were assessed: the blood-to-tissue Because the Doppler angle between the flow and the beam
ratio (BTR) in the power Doppler imaging mode and the direction is known (18°), angle correction is performed to
mean squared velocity error (MSE). The BTR is defined derive the flow velocity from its projection along the direc-
as the average power signal within the flow vessel divided tion of the Doppler beam. The estimated MSE is found to
by the average power in surrounding tissues. The mean be very similar for both methods: 0.25 and 0.27 cm/s for
squared velocity error (MSE) corresponds to the devia- conventional color flow imaging and ultrafast plane wave
tion of the experimental flow pattern from a theoretical compounding, respectively.
Poiseuille flow pattern distribution. The performance of conventional method and plane
The measured BTR values are very similar for both im- wave compounding for color flow imaging is summarized
aging sequences: 17.7 and 17.8 dB for conventional color in Table III. Quantitative values are provided for NAngles
TABLE II. P   C   F D W  U
C M.
Conventional Compounded
focused plane waves
Depth (mm) 50 50
PRFmax (kHz) 14 14
Lines in segments 9 —
Angles — 9
PRFflow (kHz) 1.55 1.55
Lines to image 63 63
Number of segments 7 —
Ensemble length 11 11
Number of firings 693 99
Acquisition time (ms) 49 7
All parameters are identical except that the acquisition time is 7 times faster in the compound method.

7. 140 IEEE TRANSACTIONS ON ULTRASONICS, FERROELECTRICS, AND FREQUENCY CONTROL , . 58, . 1, JANUARY 2011
TABLE III. O   I P  E M.
Conventional Flat Compound Compound
focused (NAngles = 1) (NAngles = 9) (NAngles = 16)
Axial res. (mm) 1.07 1.10 1.01 1.02
Lateral res. (mm) 0.54 0.86 0.53 0.53
Contrast (dB) −42 −23 −37 −40
Mean SNR gain (dB) 0 −7 +2.5 +5
BTR (dB) 17.7 12 17.8 19.1
MSE 0.25 0.34 0.27 0.27
Frame rate max (Hz) 100 10000 1600 800
The frame rate is calculated for a typical 4-cm-deep image comprising 128 lines. Compared with conventional
color flow imaging, the plane wave compounding method reaches similar performances but exhibits a frame rate
8 to 16 times higher.
= 1, NAngles = 9, and NAngles = 16 angles and compared TABLE IV. P   C E
with the conventional color flow imaging. One can observe B  F  C D.
that most values are similar, except the maximal PRF is 8 Focused Compound
to 16 times faster for the ultrafast compounding approach Depth (mm) 25 25
depending on the number of angles used (9 or 16). PRFmax (kHz) 24 24
Reducing the plane wave compounding sequence to Lines in segment 8 —
Angles — 8
NAngles = 9 angles enables the generation of a very fast im- PRFflow (kHz) 3 3
aging modality (16 times faster than the focused one) with Lines to image 128 128
a very moderate loss in contrast (5 dB) and equivalent res- Number of segments 16 —
olution and SNR compared with the conventional mode. Number of frames 11 176
Moving to the 16-angle sequence drops the maximum flow Total number of firings 1408 1408
Acquisition time (ms) 58 58
velocity detectable and the gain in acquisition time by a
factor of 2 but allows contrast equivalent to conventional All parameters are identical except for the number of images.
color flow. Trade-offs are therefore possible depending on
the type of flow analyzed. The next two sections illustrate
how the optimization of such ultrafast sequences could TABLE V. A P  I (API) P
  T M.
pave the way to many potential improvements and new
features of Doppler flow imaging. In Section IV, a method Negative
peak pressure Ispta
for the improvement of color flow imaging is presented and
Imaging method (MPa) MI (mW/cm2)
tested in vivo. In Section V, a new way to acquire, process,
Conventional focused 2.46 1.1 < 1.9 220 < 720
and analyze Doppler information with the use of ultrafast
Ultrafast compound 1.79 0.8 < 1.9 120 < 720
compounded sequences is introduced.
IV. I C F I: I V 1) Experiment Setup: Experiments were performed in
E vivo on the common carotid of a healthy volunteer imaged
using an 8-MHz linear probe (256 elements). To compare
In Section III, it was shown that the performance of both methods at the same instant in the cardiac cycle, the
conventional color flow imaging can be reached in much ultrasonic sequence consists of 58 ms of conventional fo-
shorter acquisition time. Section IV investigates the pos- cused imaging, immediately followed by 58 ms of ultrafast
sibility of using the resulting available time to introduce plane wave compounding. The parameters for the acquisi-
additional color flow improvements. The sequences are tions are shown in Table IV. The number of firings, PRFs,
optimized in a different manner depending on the appli- and duration of the acquisition are exactly the same for
cation considered: for fast-flow optimization, steep wall both sequences. Only the total number of final images is
filtering and fast acquisition times are required, whereas different: the ultrafast plane wave compounding sequence
for low-flow optimizations, very high sensitivity and reso- is able to acquire 176 frames, whereas the focused method
lution are pursued. generates only 11 frames because of the line-by-line ac-
quisition.
A. Fast Flow Optimization To ensure patient safety, acoustic power and intensity
(API) measurements were performed for both modes. The
To improve fast flow imaging, the number of angles results are shown in Table V. Measurements have been
must be limited to keep the PRF high enough. Thus, the performed following 60601–2-37 international guidance
ensemble length can be increased to fit the same acquisi- [21]. Values for both modes are below FDA safety limita-
tion time as for the conventional focused method. tions and plane wave compounding is found to outperform

8.   .:   D  141
Fig. 7. Thyroid scans using (a) ultrafast compound Doppler imaging and
(b) focused. Horizontal lines of the Doppler images are shown on (c) and
(d) for the ultrafast compound Doppler image and (e) and (f) for the
focused image.
Fig. 6. Focused color Doppler image (a) and compound color Doppler (b) for the fastest sequence (corresponding to an ensemble
using the same acquisition time. The compound image has a very low
variance and is of much better quality than the focused one. (c) to (f)
length equal to 12) the ultrafast plane wave compounding
are computed by reducing the number of frames to calculate the ultrafast outperforms the quality of the conventional method. It is
Doppler image, the acquisition time is accelerated by a factor of 2 to 14. important to note that the apparent change in the granu-
The image accelerated 14 times has a similar quality than the standard larity between the fastest ultrafast compound image [Fig.
focused image. No spatial or averaging filter is used in all these images. 6(f)] and the focused image [Fig. 6(a)] comes from the fact
that the conventional focused image is built gradually on a
segment-by-segment basis (16 segments, each containing 8
conventional color flow in terms of mechanical index (MI) lines). Thus, the spatio-temporal continuity is not ensured
and spatial peak time average intensity (Ispta). from one segment to the next, whereas the whole image is
acquired simultaneously in the compound approach.
2) Improving Velocity Measurement Accuracy: Figs.
6(a) and 6(b) compares the images obtained with the two B. Low-Velocity Flow Optimization
methods. These images are the direct output of the Dop-
pler frequency calculated without any kind of spatial or For clinical applications where low-velocity flow must
temporal smoothing. As one can observe, the ultrafast be detected in small vessels, the image contrast becomes
plane wave compound Doppler image reaches a very high a key parameter. Here, the number of angles used to com-
quality because of the long ensemble used (176 firings) pute a full color image is increased to 16 to obtain con-
whereas the conventional Doppler method generates an trast performance similar to the conventional approach
image with a high variance because of the limited tempo- (see Fig. 3) and a higher SNR (see Fig. 4). The spatial
ral averaging offered by the ensemble of just 11 firings. resolution is also explicitly increased by choosing large
Using the same set of data, one interesting experiment maximum tilting angles (±12° instead of ±9°). Both meth-
is to progressively reduce the number of acquisition frames ods (conventional and ultrafast compound) are evaluated
in ultrafast plane wave compounding (i.e., the ensemble on the thyroid of a healthy volunteer and presented in
length) to increase the temporal resolution at the expense Fig. 7.
of image quality. Figs. 6(c) to 6(f) present results from The ultrafast compound image exhibits higher flow sen-
such an experiment, with ultrafast Doppler images formed sitivity and less variance (14 times less) than the focused
using ensemble lengths of 88, 44, 22, and 12 firings. Com- one; for example, small vessels deep in thyroid are clearly
pared with the conventional Doppler method, the acqui- resolved, whereas they remain very difficult to detect in
sition time is reduced respectively by a factor 2, 4, 8, the focused image. This is demonstrated in Figs. 7(c)–(f),
and 14. Although the ultrafast Doppler image progres- where the Doppler intensity is plotted along two horizon-
sively degrades with the reduced number of acquisition tal lines (2- and 2.2-cm depth) for both methods. Peaks of
frames, it remains better than conventional Doppler. Even Doppler signals are clearly present on the ultrafast com-

9. 142 IEEE TRANSACTIONS ON ULTRASONICS, FERROELECTRICS, AND FREQUENCY CONTROL , . 58, . 1, JANUARY 2011
rospective manner. To assess the performance of the new
acquisition scheme, experiments were performed on the
carotid of a healthy volunteer.
The implemented sequence is an ultrafast sequence
comprising 3 angles (−3°, 0 °, +3°) and an acquisition PRF
between angles equal to 20 kHz. The acquisition contains
gaps between sets of compounded images to obtain a flow
acquisition frame rate of 3 kHz, instead of the maximum
flow PRF of 6.66 kHz (20 kHz/3 angles). Doppler data
are acquired at this frame rate during a complete cardiac
cycle (3000 images in total).
Fig. 8(a) shows an example of the raw in-phase/quadra-
ture-phase (IQ) signal (The IQ signal corresponds to the
beamformed and demodulated ultrasound signal) at one
given location inside the arterial blood flow. A wall filter
is applied to this raw signal to extract the blood flow sig-
nal, which is displayed on Fig. 8(b). The mean Doppler
frequency is then calculated using 15 samples in a slid-
ing window (with a 10-sample step between adjacent win-
dows). The resulting Doppler frequency signal is shown in
Fig. 8. Example of the signals in an ultrafast acquisition. (a) Signal I Fig. 8(c) and has a temporal resolution of 150 Hz, which
at a selected point in the artery in arbitrary units. (b) After applying
is significantly higher than for the standard focused color
the wall filter, the raw signal corresponding to blood flow is extracted.
(c) The mean Doppler frequency is calculated using a 15-point sliding Doppler imaging (typically 10 to 15 Hz).
window for Fourier analysis. The complete movie comprises as much as 300 images.
Fig. 9 presents some interesting frames from the acquisi-
tion. To identify the temporal location within the cardiac
pound image, whereas the same peaks are very close to cycle, Fig. 9(a) shows the mean Doppler frequency of the
the noise background on the focused one. flow in the artery. In Fig. 9(b), the flow velocity is at its
The ultrafast compound method offers much better flow minimal value. In Figs. 9(c) and 9(d), the aortic valve
detection and higher sensitivity and resolution than the opens and the flow accelerates. Between time steps (d)
conventional method. Moreover, the penetration is signifi- and (f), the flow dynamics become more complex and
cantly higher in the ultrafast compound method, mainly the Reynolds number can reach the critical value where
because of a better SNR at greater depths. turbulence can appear [22]. In Figs. 9(e.1) to (e.5), a
sequence of 5 frames shows the rapid inversion of the
V. U D: F C parabolic laminar flow into a profile where the speed is
 F F temporarily almost zero in the center of the artery. In
Fig. 9(f), local turbulent phenomena start to develop.
Standard color Doppler is limited to frame rates of up The spatio-temporal trajectory and evolution of this lo-
to 20 to 30 Hz. At such frame rates, many fast transient cally turbulent flow can be followed during a few frames.
phenomena such as turbulence or short duration flow re- Finally, in Figs. 9(g) and 9(h), the profile becomes lami-
versals are invisible. Therefore, color flow imaging could nar again.
tremendously benefit from higher acquisition rates. The two cineloops corresponding to Figs. 9(e.1) to (e.5)
Because the acquisition time of the ultrafast compound and Figs. 9 (f.1) to (f.5) have a total duration of 40 ms,
Doppler images is significantly shorter than for the con- which is the required time to perform only a single Dop-
ventional method, sequences can be designed to increase pler image in conventional color flow. This shows that such
the temporal sampling rate of the Doppler data. In this temporal close-ups as the ones allowed by the ultrafast
section, we investigate the potential of the ultrafast com- compound imaging method could be of great interest to
pound Doppler method to acquire Doppler data at high fully analyze complex flow dynamics in arteries.
spatial and temporal sampling rates.
B. Full Spectra Analysis
A. Ultrafast Doppler
The continuous acquisition of ultrafast compound im-
In this particular acquisition mode, Doppler data over aging over a 2-D area of interest offers additional possi-
the full ROI is acquired at high repetition frequency (typi- bilities for advanced flow quantification. For example, the
cally the PRF used in pulsed-wave Doppler mode) dur- acquired data are perfectly suitable for generating full-
ing a complete cardiac cycle. The highly sampled Doppler spectral analysis Doppler sonograms as those obtained by
data are then stored into memory and are available for the standard PW Doppler mode simultaneously for each
multiple parallel processing and analysis schemes in a ret- pixel of the 2-D color flow image. Therefore, we can per-

10.   .:   D  143
Fig. 9. Some selected frames of a complete cardiac cycle obtained with the ultrafast compound. (a) Average flow in the artery indicating the selected
frames. (b) Before the opening of the aortic valve, there is a minimal laminar flow. (c) and (d) Acceleration of the flow. (e) Inversion of the parabolic
profile in the deceleration. (f) Local turbulence is present and propagates in the artery. (g) and (h) Laminar flows in diastole.
form retrospective full-spectral analysis at arbitrary mul- that the acquisition parameters corresponding to the in-
tiple points throughout the whole area of interest, unlike put data of this figure are the same as those specified in
conventional PW Doppler which is restricted to a given the previous paragraph.
location. The obtained Doppler spectra for each desired By offering the possibility of performing full spectral
pixel exhibit perfect time alignment, because they are analysis throughout the color image, plane wave com-
based on data acquired at the same time. Fig. 10 shows pound sequences enable generation of Doppler sonograms
two such Doppler spectra, obtained from sample volumes of higher dimensionality (Doppler frequency + time +
denoted as b and c in the ultrafast Doppler image. Note depth + lateral position) which can be exploited in a va-
that the vertical elongation which is visible in some parts riety of ways. For example, Fig. 11 shows a new type of
of the spectra in Fig. 10 is a direct consequence of ap- flow analysis by defining longitudinal and transverse lines
plying a rectangular windowing function to the Doppler within the ultrafast color flow image, and generating Dop-
time sequence before the fast Fourier transform. Also note pler spectrum sonograms versus spatial position at mul-

11. 144 IEEE TRANSACTIONS ON ULTRASONICS, FERROELECTRICS, AND FREQUENCY CONTROL , . 58, . 1, JANUARY 2011
also be envisioned to improve the efficiency of the Doppler
exam workflow. For example, the location of maximum
peak velocity over the cardiac cycle could be automatical-
ly detected and the full Doppler spectrum corresponding
to this specific location can be calculated and displayed.
More generally, this new Doppler sequencing opens the
possibility of complete offline quantitative analysis of the
blood flow from a single ultrafast compound acquisition.
VI. D
This paper introduces a new approach of Doppler blood
flow imaging that significantly outperforms conventional
Doppler imaging. This approach is based on the successive
transmissions of compounded plane waves with different
tilting angles. The backscattered echoes corresponding to
these compounded plane waves transmissions are recom-
bined coherently to resynthesize ultrasonic images that ex-
hibit excellent contrast and highly improved frame rates.
Thanks to plane wave insonifications, each pixel of the
Fig. 10. (a) Two sample volumes plus (b) and (c) the corresponding imaged area can be formed using many more time samples
spectrograms using IQ data acquired at the same time. than those used in conventional Doppler imaging.
As an immediate consequence, several key improve-
ments can be obtained relative to conventional Doppler
tiple times. Negative flows are clearly quantified at the imaging. First of all, the acquisition time can be great-
time of the turbulence, Fig. 11(b). ly reduced, typically by a factor of up to 16 (with the
This kind of spatial analysis could eventually be done 9-angle sequence). Therefore, plane wave compounding
by allowing a user to manually draw a line to analyze a is extremely convenient for fast-flow imaging (arteries,
particular trajectory of the flow. Thanks to the ultrafast large veins) where transitory and turbulent flows could
plane wave compounding technique, automation tools can be imaged with a much better temporal resolution. We
Fig. 11. Doppler frequency versus spatial position spectrograms at three different times (a), (b), and (c). Spatial analysis is done longitudinally (x)
and transversely (z) through the common carotid artery.

12.   .:   D  145
demonstrated higher-quality flow estimation for fast- a large region of interest with excellent spatio-temporal
speed flows, as shown in Section IV, without the need for continuity.
post-processing operators (spatial smoothing, temporal Of course, the possibility of implementing such a mode
smoothing, etc.) which are typically used in conventional in real time on an ultrasound system is also a real chal-
Doppler flow imaging to improve the quality of the raw lenge, because it requires a complete redefinition of the ul-
flow estimates. On the other hand, the technique is also trasonic system architecture. The two main requirements
extremely interesting for slow-flow imaging, where resolu- are:
tion, sensitivity, and contrast are important. In that case,
the angle range and the number of angles are increased Highly parallelized acquisition and processing ca-
(from 9 to 16 as illustrated in Section IV) at the expense pabilities to perform ultrafast imaging sequences,
of a lower gain in acquisition time. Flow detection and i.e., imaging of the medium at ultrafast frame rates
definition can be strongly improved for low-speed flows, as (>1000 Hz),
shown in Fig. 7: because each pixel is derived from many Very high processing performance to be able to gener-
more time samples, much better sensitivity is enabled. Ul- ate ultrafast color images in real time. Overall, pro-
trafast compound Doppler imaging could provide a way to cessing capabilities of the ultrasound device need to
image very-low-speed flows in small vessels; for example, be increased at least by a factor of NAngles compared
for evaluating tumor recurrence after chemotherapy or ra- with conventional techniques (each pixel is beam-
diotherapy treatments (prostate or breast cancer) [23]. formed NAngles times) to provide real-time Doppler
Complex flows are difficult to analyze because of the information.
inability of conventional approaches to obtain Doppler
spectra simultaneously at several locations. To partly The requirements of such an ultrasonic platform can only
overcome these limits, Tortoli et al. proposed multigat- be provided by a highly flexible software architecture,
ed Doppler spectrum analysis to increase the amount of in which the system is able to acquire and process ul-
quantitative information available compared with conven- trasound images at ultrafast frame rates. A first clinical
tional Doppler [24], [25]. A highly-relevant feature of ul- system based on a fully software-based architecture was
trafast Doppler in this context is its ability to fully over- developed in the framework of shear wave imaging (Aix-
come the fundamental limitation of conventional Doppler, plorer ultrasound system, SuperSonic Imagine). Ultrafast
which is able to assess blood flow simultaneously only in data are stored in a digital memory and transferred at
a very limited area of interest (this refers both to color 3.5 Gbytes/s via a PCI express link to a software-based
flow imaging and PW color modes). In ultrafast Doppler, processing block that leverages GPU processing power (1
blood flow can be estimated simultaneously at each pixel CPU with 6 cores, 3.33 GHz sampling, and 1 GPU board).
location in a wide 2-D region of interest. Because blood All image lines can therefore be processed in parallel, en-
flow is evaluated simultaneously for all pixels, it enables abling generation of a few thousands of ultrasound images
a much better understanding of complex flows. Assessing per second. The Aixplorer system is currently leveraged
the full spatial and temporal distribution of flow enables, for integrating real-time ultrafast Doppler imaging. Tech-
for example, tracking pf turbulent flows, or the study of nical and clinical performances will be presented in a fu-
viscosity or Reynold’s number thanks to the dynamics of ture work.
velocity distribution profiles in the artery. It also enables
the analysis of complex flow trajectories and dynamics
because the complete Doppler spectrum becomes available VII. C
for each pixel. One should notice here that the maximum
detectable frequency/velocity is NAngles times lower than By moving beyond the concept of conventional ultrason-
in classic PW single mode. NAngles should therefore be set ic imaging acquisitions, ultrafast plane wave compounding
as low as possible for fast-flow analysis. The concept of enables revision of Doppler imaging and paves the way to
plane wave compounding can also easily be extended to new perspectives in Doppler flow analysis. First, a signifi-
transverse Doppler measurements by using independent cant gain in acquisition time (up to 16 times faster) can
sub-apertures of the array [26]. be achieved while keeping the Doppler mode performance
Finally, plane wave compounding provides a much similar to today standards, offering significant frame rate
better use of the color and PW Doppler modes, because improvements and the opportunity for improved visualiza-
they are fully integrated within the same real-time mode. tion and advanced analysis of transient flow phenomena
Perhaps more importantly, this new concept of ultrasonic such as turbulence and jets. Second, the ultrafast Doppler
sequences also provides the possibility of obtaining quanti- sequence can be optimized so that it offers comparable
fiable Doppler spectra at all image locations in a retrospec- frame rates to conventional Doppler flow imaging but with
tive manner, by using previously stored ultrafast Doppler much longer ensembles, to enhance flow imaging perfor-
data. More generally, ultrafast compound Doppler imaging mance by increasing resolution, sensitivity, and introduc-
opens up the possibility of exploring advanced flow imag- ing very fine tissue/flow discrimination. Finally, ultrafast
ing and quantification techniques by taking advantage of Doppler imaging can be used to acquire Doppler informa-
the simultaneous acquisition of Doppler information over tion at very high spatial and temporal sampling rates,

13. 146 IEEE TRANSACTIONS ON ULTRASONICS, FERROELECTRICS, AND FREQUENCY CONTROL , . 58, . 1, JANUARY 2011
allowing full spectral analysis on a large 2-D ROI in real- [20] T. Loupas, J. T. Powers, and R. W. Gill, “An axial velocity estima-
tor for ultrasound blood flow imaging, based on a full evaluation of
time as well as using previously stored data. Therefore the Doppler equation by means of a two-dimensional autocorrela-
this approach has the potential to strongly improve all tion approach,” IEEE Trans. Ultrason. Ferroelectr. Freq. Control,
aspects of currently used flow imaging and analysis appli- vol. 42, no. 4, pp. 672–688, 1995.
[21] Particular Requirements for the Safety of Ultrasound medical Di-
cations, as well as to expand the clinical applications and agnostic and Monitoring Equipment, international standard IEC
diagnostic capabilities of Doppler ultrasound far beyond 60601-2-37 2nd ed., 2007.
what is currently available. [22] Y. C. Fung, Biomechanics, Circulation, 2nd ed., New York, NY:
Springer Science, 1997, pp. 136–138.
[23] O. Rouviere, T. Vitry, and D. Lyonnet, “Imaging of prostate cancer
local recurrences: Why and how?” Eur. Radiol., vol. 20, no. 5, pp.
R 1254–1266, May 2010.
[24] P. Tortoli, V. Michelassi, and G. Bambi, “Interaction between
[1] D. H. Evans, W. N. McDicken, R. Skidmore, and J. P. Woodcock, secondary velocities, flow pulsation and vessel morphology in the
Doppler Ultrasound, Physics, Instrumentation, and Clinical Applica- common carotid artery,” Ultrasound Med. Biol., vol. 29, no. 3, pp.
tions. New York, NY: Wiley, 1989. 407–415, 2003.
[2] J. A. Jensen, Estimation of Blood Velocities Using Ultrasound: A [25] W. Secomski, A. Nowicki, P. Tortoli, and R. Olszewski, “Multigate
Signal Processing Approach. New York, NY: Cambridge University Doppler measurements of ultrasonic attenuation and blood hemat-
Press, 1996. ocrit in human arteries,” Ultrasound Med. Biol., vol. 35, no. 2, pp.
[3] L. Y. L. Mo, T. L. Ji, C. H. Chou, D. Napolitano, G. W. McLaugh- 230–236, Feb 2009.
lin, and D. DeBusschere, “Zone-based color flow imaging,” in Proc. [26] R. Daigle, L. Pflugrath, J. Flynn, K. Linkhart, and P. Kaczkowski,
IEEE Ultrasonics Symp., 2003, pp. 29–32. “High frame rate quantitative Doppler imaging,” presented at IEEE
[4] N. Oddershede, F. Gran, and J. A. Jensen, “Multi-frequency encod- Ultrasonics Symp., Rome, Italy, 2009.
ing for fast color flow or quadroplex imaging,” IEEE Trans. Ultrason.
Ferroelectr. Freq. Control, vol. 55, no. 4, pp. 778–786, Apr. 2008.
[5] T. X. Misaridis and J. A Jensen “Space-time encoding for high
frame rate ultrasound imaging,” Ultrasonics, vol. 40, no. 1–8, pp. Jeremy Bercoff was born 1977 in Paris, France.
593–597, May 2002. In 2001, he received an engineering degree from the
[6] K. L. Gammelmark and J. A. Jensen, “Multielement synthetic Ecole Supérieure de Physique et de Chimie de Par-
transmit aperture imaging using temporal encoding,” IEEE Trans. is (ESPCI, ParisTech) with a specialization in
Med. Imaging, vol. 22, no. 4, pp. 552–563, 2003. Physics. In 2004, Jeremy received a Ph.D. degree in
[7] L. Germont-Rouet, T. Loupas, and O. Bonnefous, “Clutter filtering physics (acoustics) from the University of Paris VII
with small ensemble length in ultrasound imaging,” U.S. Patent ap- for his work on ultrafast imaging and shear-wave-
plication, US 2007112269, May 17, 2007. based elastography in soft tissues for cancer detec-
[8] D. N. Roundhill, T. Loupas, A. Criton, and D. Rust, “Coincident tion. In 2005, he co-founded SuperSonic Imagine, a
tissue and motion ultrasonic diagnostic imaging,” U.S. Patent French medical ultrasound imaging and therapy
6 139 501, Dec. 14, 2000. company in Aix en Provence, for which he led the
[9] T. K. Holfort, F. Gran, and J. A. Jensen, “Minimum variance beam- introduction in 2007 of a new real time elasticity imaging mode (shear
forming for high frame-rate ultrasound imaging,” in Proc. IEEE wave elastography) on the company’ first marketed product. He is the
Ultrasonics Symp., 2007, pp. 1541–1544. author of 8 patents and more than 15 peer reviewed papers in the field of
[10] P. Tortoli, L. Bassi, E. Boni, A. Dallai, F. Guidi, and S. Ricci, medical imaging. His current research activities include ultrafast imaging,
“ULA-OP: An advanced open platform for ultrasound research,” new ultrasound beam-forming strategies, and functional ultrasound imag-
IEEE Trans. Ultrason. Ferroelectr. Freq. Control, vol. 56, no. 10, pp. ing, such as color Doppler and elasticity imaging.
2207–2216, 2009.
[11] M. Tanter, J. Bercoff, L. Sandrin, and M. Fink, “Ultrafast com-
pound imaging for 2-D motion vector estimation: Application to
transient elastography,” IEEE Trans. Ultrason. Ferroelectr. Freq. Gabriel Montaldo was born 1972 in Montevi-
Control, vol. 49, no. 10, pp. 1363–1374, 2002. deo, Uruguay. He received the Ph.D. degree in
[12] J. Bercoff, M. Tanter, and M. Fink, “Supersonic shear imaging: A new physics from the Universidad de la Republica,
technique for soft tissues elasticity mapping,” IEEE Trans. Ultrason. Uruguay, in 2001. Since 2002, he has worked in
Ferroelectr. Freq. Control, vol. 51, no. 4, pp. 396–409, Apr. 2004. the Laboratory Ondes et Acoustique at the Ecole
[13] J. Udesen, F. Gran, K. Lindskov Hansen, and J. A. Jensen, “High Supérieure de Physique et de Chimie Industrielles
frame-rate blood vector velocity imaging using plane waves: simula- de la Ville de Paris (ESPCI). His principal re-
tions and preliminary experiments,” IEEE Trans. Ultrason. Ferro- search includes applications of the time-reversal
electr. Freq. Control, vol. 55, no. 8, pp. 1729–1743, Aug. 2008. process to adaptive focusing in heterogeneous me-
[14] J. Y. Lu, “2D and 3D high frame rate imaging with limited diffrac- dia, biomedical applications of ultrasound, and
tion beams,” IEEE Trans. Ultrason. Ferroelectr. Freq. Control, vol. elastography. He has published more than 30 re-
44, no. 4, pp. 839–855, 1997. viewed articles in the ultrasound domain.
[15] G. Montaldo, M. Tanter, J. Bercoff, N. Benech, and M. Fink,
“Coherent plane-wave compounding for very high frame rate ul-
trasonography and transient elastography,” IEEE Trans. Ultrason.
Ferroelectr. Freq. Control, vol. 56, no. 3, pp. 489–506, Mar. 2009. Thanasis Loupas received the B.E. degree in
[16] D. Shattuck, M. Weinshenker, S. Smith, and O. Von Ramm, “Ex- electrical engineering from the National Technical
plososcan: A parallel processing technique for high speed ultrasound University of Athens, Greece, in 1983. He was a
imaging with linear phased arrays,” J. Acoust. Soc. Am., vol. 75, no. post-graduate research student at the National
4, pp. 1273–1282, Apr. 1984. Research Center “Democritos” in Athens, Greece
[17] S. Park, S. R. Aglyamov, and S. Y. Emelianov, “Elasticity imaging (1983–1984), and then moved to the Department
using conventional and high-frame rate,” IEEE Trans. Ultrason. Fer- of Medical Physics and Medical Engineering, Uni-
roelectr. Freq. Control, vol. 54, no. 11, pp. 2246–2256, 2007. versity of Edinburgh, Scotland, as a Ph.D. student
[18] C. Kasai, K. Namekawa, A. Koyano, and R. Omoto, “Real-time two- working on adaptive image processing algorithms
dimensional blood flow imaging using an autocorrelation technique,” for speckle reduction in medical ultrasonic imag-
IEEE Trans. Ultrason. Ferroelectr. Freq. Control, vol. 32, no. 3, pp. ing. After receiving the Ph.D. degree in 1988 from
458–463, 1985. the University of Edinburgh, he became a Research Fellow at the same
[19] O. Bonnefous and P. Pesque, “Time domain formulation of pulse- university working on real-time image processing and Doppler signal
Doppler ultrasound and blood velocity estimation by cross-correla- processing techniques. In 1992, he joined the Ultrasonics Laboratory of
tion,” Ultrason. Imaging, vol. 8, no. 2, pp. 73–85, 1986. the Commonwealth Scientific Industrial & Research Organization, Syd-

14.   .:   D  147
ney Australia, where he worked as a Research Scientist, Senior and Prin- de la Ville de Paris (ESPCI), Paris, France, and at Paris 7 University
cipal Research Scientist in novel flow and tissue-motion estimation tech- (Denis Diderot), France. In 1990, he founded the Laboratory Ondes et
niques, and quantitative analysis of cardiac and obstetrical ultrasound Acoustique at ESPCI. In 2002, he was elected to the French Academy of
data. In 1996, he became a member of the Corporate Technical Staff at Engineering, in 2003 to the French Academy of Science, and in 2008 to
Advanced Technology Laboratories, Bothell, WA, (subsequently acquired the Chair of Technological Innovation for the College de France.
by and integrated within Philips Medical Systems) where he worked on His current research interests include medical ultrasonic imaging, ul-
a variety of R&D projects in the areas of Color/PW Doppler system trasonic therapy; nondestructive testing; underwater acoustics; telecom-
architecture and processing, as well as algorithms for computer-assisted munications; seismology; active control of sound and vibration; analogies
quantification and system automation, implemented in the HDI5000 and between optics, quantum mechanics, and acoustics; wave coherence in
iU22 ultrasound systems and the QLAB quantification software. He multiply scattering media; and time-reversal in physics.
joined SuperSonic Imagine, Aix-en-Provence, France as a Principal Sci- He has developed different techniques in acoustic imaging (transient
entist in 2007, where he continues to work in the design and development elastography, supersonic shear imaging), wave focusing in inhomoge-
of signal and image processing algorithms for all imaging modes of the neous media (time-reversal mirrors), speckle reduction, and in ultrasonic
Aixplorer ultrasound system, plus automatic optimization and quantifi- laser generation. He holds more than 50 patents, and he has published
cation techniques. more than 300 articles. 4 start-up companies have been created from
his research (Echosens, Sensitive Object, Supersonic Imagine, and Time
Reversal Communications).
David Savéry earned the M.S. degree in 1998
from Mines Paris Tech, France, majoring in image
analysis and image processing.He then joined the Mickaël Tanter is a Research Professor of the
Laboratory of Biorheology and Medical Ultrason- French National Institute for Health and Medical
ics, University of Montréal Hospital, Montréal, Research (INSERM). For five years, he has been
PQ, where he received the Ph.D. degree in bio- heading the team Inserm ERL U979 “Wave Phys-
medical engineering in 2003. From 2004 through ics for Medicine” at Langevin Institute, ESPCI
2005, Dr. Savéry was a postdoctoral fellow and ParisTech, France. In 1999, he was awarded the
then senior member of the research staff at Philips Ph.D. degree from Paris VII University in phys-
Research North America, Briarcliff Manor, NY. ics.
David Savéry is currently R&D ultrasound engi- His main activities are centered on the devel-
neer at SuperSonic Imagine, Aix-en-Provence, France. He has contrib- opment of new approaches in wave physics for
uted to the development of the Aixplorer ultrasound system since 2006. medical imaging and therapy. His current research
His research interests are in statistical signal processing, medical imag- interests cover a wide range of topics: elastography using shear wave
ing, ultrasound tissue characterization, and biomechanics. imaging, high intensity focused ultrasound, ultrasonic imaging using ul-
trafast ultrasound scanners, adaptive beamforming, and the combination
of ultrasound with optics and MRI. In 2009, he received the Frederic
Lizzi Early Career Award of the International Society of Therapeutic
Fabien Meziere was born in December 1987 in Ultrasound and the Montgolfier Prize of the National Society for Indus-
Bernay, France. In 2007 he entered the Ecole Su- try Valorization (SEIN) in 2010. Mickael Tanter is the recipient of 17
périeure de Physique et de Chimie Industrielle de patents in the field of ultrasound imaging and the author of more than
Paris (ESPCI ParisTech) and in 2011 obtained an 80 technical peer reviewed papers and book chapters. He is an Associate
engineer degree with a specialization in physics. In Editor and TPC member of IEEE Ultrasonics and member of the Brain
2009, he worked at SuperSonic Imagine on ultra- Advisory board of the Focused Ultrasound Surgery Foundation. In 2005,
fast compound Doppler imaging. He is now achiev- he, along with M. Fink, J. Souquet, C. Cohen-Bacrie and J. Bercoff
ing his master’s degree in acoustics at the Univer- founded SuperSonic Imagine, an innovative French company positioned
sity of Paris VII. in the field of medical ultrasound imaging and therapy; in 2009 they
launched a new-generation ultrasound imaging platform called Aixplorer
with a unique shear wave imaging modality.
Mathias A. Fink received the M.S. degree in
mathematics from Paris University, France, in
1967, and the Ph.D. degree in solid-state physics
in 1970. Then he moved to medical imaging and
received the Doctorat es-Sciences degree in 1978
from Paris University. His Doctorat es-Sciences
research was in the area of ultrasonic focusing
with transducer arrays for real-time medical imag-
ing.
Dr. Fink is a professor of physics at the Ecole
Superieure de Physique et de Chimie Industrielles